Abstract: To improve the generalization performance of detection window extraction on people detection, a fast extraction algorithm based on depth information was proposed. The geometric centers of detection windows were obtained by calculating extreme values of frequency of depth data. And the sizes of detection windows were obtained based on the relationship between depth and height in depth image. Finally, a series of people detection windows were extracted fast. And the algorithm was evaluated from the number of detection windows, response time and the accuracy. Experimental results show the feasibility and efficiency of this method.